Determinants of Personal Exposure to Some Carcinogenic Substances and Nitrogen Dioxide Among the General Population in Five Swed

Home , Butadiene

ORIGINAL ARTICLE Determinants of personal exposure to some carcinogenic substances and nitrogen dioxide among the general population in ﬁve Swedish cities

Annika Hagenbjo¨ rk-Gustafsson1, Andreas Tornevi1, Eva M. Andersson2, Sandra Johannesson2, Tom Bellander3, Anne-Sophie Merritt3, Ha˚kan Tinnerberg4,Ha˚kan Westberg5,6, Bertil Forsberg1 and Gerd Sallsten2

Environmental levels of airborne carcinogenic and related substances are comparatively better known than individual exposure and its determinants. We report on a personal monitoring program involving five Swedish urban populations. The aim of the program was to investigate personal exposure to benzene, 1,3-butadiene, formaldehyde, and nitrogen dioxide (NO2). The measurements were performed among 40 inhabitants during seven consecutive days, in one urban area each year, during 2000–2008. The estimated population exposure levels were 1.95 mg/m3 for benzene, 0.56 mg/m3 for 1,3-butadiene, 19.4 mg/m3 for formaldehyde, 3 and 14.1 mg/m for NO2. Statistical analysis using a mixed-effects model revealed that time spent in traffic and time outdoors contributed to benzene and 1,3- butadiene exposure. For benzene, refueling a car was an additional determinant influencing the exposure level. Smoking or environmental tobacco smoke were significant determinants of exposure to NO2, benzene, and 1, 3-butadiene. Those with a gas stove had higher NO2 exposure. Living in a single-family house increased the exposure to formaldehyde significantly. In a variance component model, the between-subject variance dominated for 1,3-butadiene and formaldehyde, whereas the between-city variance dominated for NO2. For benzene, the between-subject and between-cities variances were similar.

Journal of Exposure Science and Environmental Epidemiology (2014) 24, 437–443; doi:10.1038/jes.2013.57; published online 25 September 2013 Keywords: personal exposure; benzene; 1,3-butadiene; nitrogen dioxide; formaldehyde; mixed models

INTRODUCTION indoor environments, personal activity patterns and emission from 6–8 Urban populations are exposed to complex mixtures of air indoor sources. The relation between these estimates and pollutants. A large part of these compounds originate from measured individual exposure may thus be weak and therefore 9 combustion, for example, in motorized traffic, and some of these add exposure misclassification. An important alternative method are known carcinogens.1 of estimating population exposure is therefore to perform personal Ideally, epidemiological studies of carcinogenic effects of urban measurements. Personal measurements could also serve as an air pollution would be based on life-time individual inhalation important tool for validation of exposure assessment made by a exposure to all carcinogenic (and co-carcinogenic) compounds in combination of models and geographical information. urban air. This is practically not feasible. Population-based studies, Consequently, the Swedish Environmental Protection Agency therefore, typically rely on estimates of levels of specific compounds has taken an initiative to collect data on personal exposure for in selected locations. Compounds that are commonly chosen for some common air pollutants: benzene, 1,3-butadiene, formalde- such studies include, for example, benzene, 1,3-butadiene, and hyde, and NO2. The survey is conducted every year according to a formaldehyde2 as well as other compounds or metrics that relate to rotating study plan, in one of five selected urban areas. certain source types. The concentration of nitrogen oxides (NOx)or The aims of the present study are: nitrogen dioxide (NO2) is often used as an indicator of motor vehicle exhaust, including diesel engine exhaust, classified as To evaluate the environmental exposure of the adult popula- carcinogenic by IARC.3 Individual or population long-term exposure tion, aged 20–50 years, to benzene, 1,3-butadiene, formalde- is typically estimated by modeling of outdoor levels at place of hyde, and NO2, with regard to average levels as well as variation residence,4 either by land-use regression or dispersion modeling.5 within and between individuals from 2000 to 2008. Short-term exposure is often evaluated using fixed-site monitoring To assess the contribution of smoking, environmental tobacco data. In addition to the methodological uncertainties of these smoke (ETS), time spent in traffic, refueling of vehicles, and methods, they omit the contribution of infiltration of outdoor air to additional potential sources of personal exposure.

1Occupational and Environmental Medicine, Umea University, Umeå, Sweden; 2Department of Occupational and Environmental Medicine, Sahlgrenska University Hospital and Academy, University of Gothenburg, Box 41, Gothenburg, Sweden; 3Centre of Occupational and Environmental Medicine, Stockholm County Council and Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden; 4Department of Occupational and Environmental Medicine, Lund University, Lund, Sweden; 5Man-Technology-Environment (MTM) Research Centre, O¨ rebro University, O¨ rebro, Sweden and 6Department of Occupational and Environmental Medicine, Orebro University Hospital, Orebro, Sweden. Correspondence to: Annika Hagenbjo¨rk Gustafsson, Occupational and Environmental Medicine, Umeå University, Umeå SE-901 87, Sweden. Tel.: þ 46 90 785 37 82. Fax: þ 46 90 785 24 56. E-mail: [email protected]. Received 17 April 2013; revised 26 June 2013; accepted 3 July 2013; published online 25 September 2013 Determinants of personal exposure to carcinogenic substances and NO2 Hagenbjo¨rk-Gustafsson et al 438 METHODS From 2007 onwards, the Ogawa sampler (Ogawa & Company, Pompano Study Areas Beach, FL, USA) was used instead. The triethanolamine-coated collection filters were obtained from the manufacturer (Ogawa). The uptake rate is The five cities enrolled in the study were Stockholm, Gothenburg, Malmo¨, 8.60 ml/min (±17.6%), and the detection limit is 0.081 mg/m3 for 7 days of Umeå, and Lindesberg. Stockholm, Gothenburg, and Malmo¨ were chosen, 12 as they are the three largest cities in Sweden. Umeå is the largest city in measurements of NO2. For both NO2 measuring methods, one filter was the northern part of Sweden, and Lindesberg represents a small city in the used for the whole 7-day measurement. inner parts of the country with frequent wood burning (Supplementary Table 1, Supplementary Information). This study incorporates measure- Benzene and 1,3-butadiene. In the first measurement campaign (2000), ments from eight measurement campaigns (Supplementary Table 2). benzene was collected using Perkin Elmer samplers packed with 300 mg Tenax TA. The uptake rate of the sampler is 0.42 ml/min. As from the 2001 campaign, diffusive sampling of benzene and 1,3-butadiene was Study Participant Recruitment performed using Perkin Elmer tubes packed with Carbopack X, 60/80 In each city, a randomized selection of about 150 persons, aged 20–50 mesh (Supelco, Bellafonte, PA, USA), in order to also measure 1,3- years, was made from the Swedish population registry. An invitation letter butadiene.13 During sampling, the tubes were equipped with Perkin Elmer to participate in the study was sent to the selected persons; about 30–50 diffusion caps. The uptake rate is 0.59 ml/min (±12.5%) for benzene and letters were sent at a time. Persons who had moved were excluded, and 0.56 ml/min (±10.0%) for 1,3-butadien.14,15 The analysis was performed by the remaining persons were contacted by telephone. This procedure was automatic thermal desorption coupled with capillary gas chromatography, repeated until 40 persons had accepted to participate. In the campaigns in and identification with mass spectrometry.14 Stockholm 2003, Malmo¨ 2003, and Lindesberg 2005, invitation letters with a ‘‘yes’’ or ‘‘no’’ answer sheet and a return envelope were sent out Formaldehyde. During the first two measurement campaigns (2000 and together. In Stockholm, invitation letters were sent to all of the randomized 2001), formaldehyde was collected using a modified GMD 570 Series persons at the same time, and after 3 weeks, a reminder letter was sent. Formaldehyde Dosimeter badge (GMD Systems, Hendersonville, PA, The mean response rate was 59% (Gothenburg, 2000: 71%; Umeå, 2001: USA).16 The sampler consists of a polypropylene housing and two glass- 67%; Stockholm, 2003: 44%; Malmo¨, 2003: 43%; Lindesberg, 2005: 42%; fiber filters impregnated with 2,4-dinitrophenyl-hydrazine (2,4-DNPH). The Gothenburg, 2006: 60%; Umeå, 2007: 81%; Malmo¨, 2008: 66%). standard sampler with 0.7 mg of 2,4-DNPH/filter was modified by adding an extra amount of reagent (3.5 mg DNPH/filter) to the filter to make it last Participant Activities and Questionnaires for 7 days of measurements. The uptake rate of the sampler is 20.5 ml/min (±8.6%).17 Formaldehyde was analyzed by high-performance liquid A questionnaire was distributed at the start of the measurements including chromatography with UV detection at 365 nm.16 The detection limit of questions about home characteristics (type of house, heating system, the method is 0.1 mg/sample, which corresponds to a detection limit of access to a gas stove indoors, if a garage was attached to the house), 0.5 mg/m3 for 7 days of measurements.17 smoking habits, occupation, work address, commuting, means of From 2003 to 2008, the UMEx100 sampler (SKC, Eighty Four, PA, USA) conveyance, exposure to exhaust fumes or gasoline, or to vapor or was used. The uptake rate of the sampler is 28.6 ml/min (±18%; http:// solvents during work or in leisure time. During the measurement period, www.skcinc.com/prod/500-100.asp). The analysis was performed in the the participants kept a time activity diary every day to record time spent in same way as for the GMD sampler above. different microenvironments: outdoors in traffic; outdoors at workplace; Details of measurement data and chemical analysis methods are outdoors elsewhere; indoors at home; indoors at workplace; indoors summarized in Supplementary Table 3, in Supplementary Data. elsewhere. The participants also recorded time spent near smokers and time spent in homes during burning of wood or pellets. Special activities such as handling of gasoline, painting, welding, and so on were Statistical Analysis recorded. Personal exposure was measured in the five cities on either one or two occasions. Thus, we have collected a random sample of persons and a random sample of monitoring periods within persons. The repeated- Personal Measurements measurement structure allowed the variance to be split into between- Personal measurements of benzene, 1,3-butadiene, formaldehyde, and 2 2 subject variance (sB) and within-subject variance (sW). A mixed model was nitrogen dioxide were performed during seven consecutive days among used to identify determinants with a significant impact on the individual’s 40 persons in each city. The measurements were started according to a exposure to a substance: schedule, with measurements including two to eight participants per X4 week. For 20 of the participants, repeated measurements were performed ÀÁ Y ¼ ln X ¼ m þ b þ d CITY 2–6 weeks after the first measurement. hij hij Y hi h h h 1 The study was performed in one city each year. After the measurements ¼ ; ð1Þ X5 X1 Xp in all five cities were completed, the campaign started all over again. This þ j ðCITY ÃYEAR Þþ y C þ e study design makes it possible to study time trends in exposure as well as hm h m k k hij h¼1 m¼1 k¼1 differences in exposure levels between cities. The measurement campaigns were performed during autumn, except for the Stockholm where Yhij represents the exposure concentration, on the natural campaign in 2003, which was performed during spring (Supplementary logarithmic scale, for subject (individual) i, measurement j in city h, and Table 2). Ck represents the determinants of the exposure (variables from the Diffusive samplers were carried in a string around the neck, near the questionnaires and time-activity diaries). dh, jhm, and yk are regression breathing zone. The participants were instructed to wear the samplers all coefficients, representing the effects of the covariates. The campaign in day. At night when sleeping, they were told to hang the samplers close to Stockholm was planned for the fall 2002, but carried out in spring 2003 the bed. When the participants were outdoors, at work, in a car, or using and these data were used to approximate the levels in 2002. The stochastic public transportation, the monitors were worn outside the clothes. During terms bhi and ehij in model (1) were assumed to be normally distributed 2 2 rain or snowfall, they wore the samplers underneath their outer clothing. with expected value zero and variances shB and shW. Further, it was assumed that bhi and ehij are mutually independent. Different variances were allowed for each measuring campaign (year), and the assumption of Diffusive Samplers and Chemical Analysis 2 2 common variances for all campaigns (sBand sW) was tested with the use of Nitrogen dioxide. The Willems badge sampler was used to measure the deviance (the difference between the log likelihood functions of two personal exposure to nitrogen dioxide from 2001 to 2005. The badge nested models). Those variables (from the questionnaire and the time- consists of a cylinder of polystyrene with an absorption filter, Whatman activity diary) that were significant in a multivariate model (backward GF-A glass-fiber filter (Maidstone, Great Britain) impregnated with elimination, P-value limit 0.05) were considered as important determinants triethanolamine, as an adsorbent for NO2. A Teflon filter (Schleicher & of exposure. The finally selected models were evaluated by estimating the Schuell TE 38, 5 mm, Dassel, Germany) serves as a draught shield. The reduction in variance achieved by including the determinants. The uptake rate is 40 ml/min (±22%).10 After sampling, each filter was parameters of the mixed model were estimated using PROC MIXED in extracted with ultra-pure water, and nitrite ions were analyzed in a SAS (version 9.1.3). suppressed ion chromatography system with a conductivity detector. The Sampling was thus made at three different levels: by city, by person, and detection limit for 7 days of measurements is 0.11 mg/m3.11 by measurement period. In the analysis of determinants, the cities were

Journal of Exposure Science and Environmental Epidemiology (2014), 437 – 443 & 2014 Nature America, Inc. Determinants of personal exposure to carcinogenic substances and NO2 Hagenbjo¨rk-Gustafsson et al 439 assumed to be fixed effects. This means that the results can be generalized 1,3-butadiene, indicating that a highly exposed person had almost to the five cities in this study. However, if the cities are considered to be 17 times higher exposure than a low-exposed person (the ratio random effects, the results can be generalized to the population of between the 97.5th percentile and the 2.5th percentile was 16.9). Swedish cities, under the assumption of the following model Among smokers, the between-subject variance dominated for all Yhij ¼ lnðXhijÞ¼mY þ ah þ bhi þ ehij; ð2Þ substances except for NO2 (Table 1). where Yhij is the natural logarithm of the concentration measured during the jth repetition for the ith person in the hth city, and a, b, e are random 18 Determinants of Exposure and Time Trends effects for city, person, and measurement. It was assumed that ah, bhi, ehij 2 2 The tests regarding the appropriate variance structure (different or are mutually independent with expected value zero, and variancessa, sB, 2 and sW for city, person, and measurement, respectively. mY was estimated equal variances for all measuring campaigns) showed that with a weighted mean over g cities by different variances should be used for benzene, 1,3-butadiene, ÀÁ and formaldehyde, whereas for NO2, a common within-person v1uÀÁ1 þ v2u2 þ ::: þ vgug m^Y¼ ; variance could be assumed at all campaigns. v1 þ v2 þ ::: þ vg Where uh is the (weighted) mean over the subjects in city h and vh is the Benzene. Significant determinants as well as their effect on inverted variance of each u (1/Var[u ]). The variances of the three-level h h benzene exposure are presented in Table 2. Exposure to benzene model were estimated using PROC NESTED, and m^Y was estimated using PROC MIXED (with subcommand RANDOM city id (city)). The expected was significantly higher in Lindesberg and in Stockholm compared value on a natural scale (i.e., the arithmetic mean) of x was estimated with with Umeå. For those three cities with measurements in two 2 2 2 ^ m^X ¼ expðm^Y þ 0:5ðsâ þ s^B þ s^WÞ). Each fold range, R0:95, was calculated as different years (Gothenburg, Umeå, Malmo¨), there was a significant ^ 18 R0:95 ¼ exp(3.92*s^q), where q ¼ (a, B, W). decrease over time in Umeå ( À 35%, Po0.001) and in Malmo¨ ( À 42%, Po0.001). The exposure was higher for smokers, 49% (Table 2: RESULTS exp(0.398) ¼ 1.49), with a 95% confidence interval of 26–76% Estimated Exposure in the General Population from a Three-Level (exp(0.398 À 1.96*0.084); exp(0.398 þ 1.96*0.084)). Also, the expo- Analysis sure was higher for persons who had refueled gasoline (18%). The population exposure was estimated using the most recent Using the mean exposure 1.95 in Table 1, 18% corresponds to measurement campaign for each city (Stockholm 2003, Lindesberg 0.35 mg/m3. The exposure also increased with time spent outdoors: 2005, Gothenburg 2006, Umeå 2007, and Malmo¨ 2008) using if two people are compared who spend either 2% or 4% of their Eq. 2 (Table 1). time outdoors, respectively, the person spending 4% outdoors has The estimated arithmetic mean concentrations for the general 3.6% higher exposure. population in Sweden (2000–2008) were 1.95 mg/m3 for benzene, The results among non-smokers were similar to those among all 0.56 mg/m3 for 1,3-butadiene, 19.4 mg/m3 for formaldehyde, and subjects (Table 2). For non-smokers, a significant effect of time 3 14.1 mg/m for NO2. The exposure level of 1,3-butadiene was 92% spent near smokers was found (P ¼ 0.026): if two people are higher for smokers compared with non-smokers, for benzene it compared who spend either 2% or 4% of their time near smokers, was 14% higher, and for NO2 11% higher. Details on personal respectively, the person with the longer time will have 3.4% exposure are shown in Supplementary Table 2. higher exposure. An effect of traffic was also found (P ¼ 0.037): if The between-subject variance dominated for 1,3-butadiene and two people are compared who spend either 2% or 4% of their formaldehyde, whereas the between-city variance dominated for time in traffic, respectively, the person with longer time has 3.3% NO2 (all subjects). For benzene, the between-subject and higher exposure. between-cities variances were similar (the fold ranges were about The results from the mixed model for benzene (Table 2) 7). The highest between-subject fold range, 16.9, was observed for were used to estimate the personal exposure for a person with

Table 1. Estimated mean concentrations and variance components for benzene, 1,3-butadiene, formaldehyde, and NO2 from personal measurements in ﬁve Swedish cities.

2 ^ 2 ^ 2 ^ 3 Observations (subjects) s^a(R0:95) s^B(R0:95) s^W(R0:95) m^Y m^Xmg/m All Benzene 295 (195) 0.246 (6.99) 0.234 (6.66) 0.132 (4.15) 0.363 1.95 1,3-Butadiene 295 (195) 0.232 (6.61) 0.520 (16.89) 0.301 (8.59) À 1.11 0.559 Formaldehyde 292 (195) 0.104 (3.54) 0.142 (4.38) 0.057 (2.55) 2.81 19.4 a NO2 217 (146) 0.203 (5.85) 0.113 (3.73) 0.149 (4.54) 2.41 14.1

Non-smokers Benzene 249 (165) 0.272 (7.72) 0.202 (5.82) 0.142 (4.38) 0.335 1.90 1,3-Butadiene 249 (165) 0.229 (6.53) 0.449 (13.83) 0.331 (9.54) À 1.21 0.491 Formaldehyde 246 (165) 0.101 (3.48) 0.128 (4.07) 0.059 (2.59) 2.82 19.4 a NO2 185 (124) 0.174 (5.13) 0.101 (3.48) 0.163 (4.87) 2.41 13.9

Smokers Benzene 45 (29) 0.170 (5.03) 0.386 (11.42) 0.076 (2.95) 0.457 2.17 1,3-Butadiene 45 (29) 0.347 (10.07) 0.551 (18.35) 0.143 (4.40) À 0.580 0.942 Formaldehyde 45 (29) 0.080 (3.03) 0.246 (6.99) 0.048 (2.36) 2.73 18.5 a NO2 31 (21) 0.555 (18.55) 0.102 (3.50) 0.063 (2.67) 2.38 15.4

2 2 2 The estimated arithmetic mean (m^X ) and the mean of log-transformed data (m^Y ), and variance components (s^a, s^B, s^W; city, person, measurement), with their fold range (^R0:95), are presented. a No measurements of NO2 in Gothenburg.

& 2014 Nature America, Inc. Journal of Exposure Science and Environmental Epidemiology (2014), 437 – 443 Determinants of personal exposure to carcinogenic substances and NO2 Hagenbjo¨rk-Gustafsson et al 440 Table 2. Benzene exposure, estimates of ﬁxed effects, for all subjects and for non-smokers (no observations were excluded).

Variable Level All (430 observations, 291 subjects) Non-smokers (352 observations, 238 subjects)

Estimate (SE) P Estimate (SE) P

Intercept À 0.118 (0.070) 0.093 À 0.096 (0.072) 0.183 city Gothenburg À 0.285 (0.146) 0.052 À 0.282 (0.151) 0.063 Lindesberg 0.670 (0.139) o0.001 0.714 (0.150) o0.001 Malmo¨ 0.047 (0.106) 0.654 À 0.174 (0.101) 0.087 Stockholm 1.025 (0.101) o0.001 0.906 (0.090) o0.001 Umea˚0 0 City*year Gothenburg, 2000 0.209 (0.143) 0.146 0.174 (0.158) 0.273 (time trend) Umea˚, 2001 0.429 (0.124) o0.001 0.390 (0.133) 0.004 Stockholm, 2002 0 0 Malmo¨, 2003 0.548 (0.126) o0.001 0.578 (0.103) o0.0001 Lindesberg, 2005 0 0 Gothenburg, 2006 0 0 Umea˚, 2007 0 0 Malmo¨, 2008 0 0 Smoke*ETS No No 0 Not included No Yes 0.077 (0.069) 0.275 Not included Yes (No and Yes) 0.398 (0.084) o0.001 Not included ETSa (Proportion) Not included 1.685 (0.746) 0.026 Outdoora (Proportion) 1.746 (0.667) 0.001 1.900 (0.665) 0.005 Gasoline No 0 NS Yes 0.166 (0.059) 0.009 Traffica (Proportion) NS 1.606 (0.759) 0.037 Abbreviations: ETS, environmental tobacco smoke; NS, non-significant. Variables in the starting model were city, city*year, sex, smoke* ETS, type of house, occupational exposure, open fire, outdoors, traffic, home, garage, gasoline, oil heating. aProportion of time spent in specific environment. Range of proportion was 0–0.34 (outdoor), 0–0.39 (traffic), 0–0.94 (ETS).

outdoors, respectively, the person spending 4% outdoors has 4.8% Table 3. Expected exposure (mg/m3) for a subject with specified higher exposure. The exposure also increased with a higher background data, estimated from the final models using all proportion of time spent in traffic (P ¼ 0.045): if two people are observations. compared who spend either 2% or 4% of their time in traffic,

a b c d respectively, the person spending 4% will have 4.7% higher Benzene 1,3-Butadiene NO2 Formaldehyde exposure. Gothenburg, 2000 0.95 30.4 The results among non-smokers were similar to those among all Umea˚, 2001 1.73 0.44 8.7 24.0 subjects (Supplementary Table 4). A significantly higher exposure Stockholm, 2002 2.92 0.45 16.3 17.8 was found among those who had spent time near smokers (25% Malmo¨, 2003 2.02 0.44 15.4 21.6 higher, P ¼ 0.0425). Among non-smokers, the effect of time spent Lindesberg, 2005 2.37 0.54 6.5 30.8 Gothenburg, 2006 0.91 0.15 30.0 in traffic was not significant. Umea˚, 2007 1.01 0.54 13.0 21.4 In Table 3, the estimated 1,3-butadiene concentrations (using Malmo¨, 2008 1.11 0.51 14.6 25.4 the model for all subjects, Supplementary Table 4) for a non- aNon-smoker, no time in ETS, 2% outdoors, no refueling of gasoline. smoker from various cities and campaigns are presented for bNon-smoker, no time in ETS, 2% outdoors, 5% in traffic. comparison. This person is assumed to spend 2% of the time cNo occupational exposure, 70% at home, no gas stove or oil heating. outdoors and 5% of the time in traffic, but no time exposed to ETS. dNon-smoker, no time in ETS, living in single-family house.

NO2. The results for all subjects showed that the NO2 exposure was significantly lower in Lindesberg compared with Umeå, and certain activities, see Table 3. The benzene concentrations for a significantly higher in Stockholm compared with Umeå, non-smoker in Umeå 2007, who did not refuel gasoline, Supplementary Table 5. For the two cities with measurements in spent 2% of the time outdoors, and was not exposed to ETS 3 different years (Umeå and Malmo¨), a significant increase over time was estimated to be 1.01 mg/m ( exp( À 0.083 þ 0.18/2) ¼ 1.01, was found in Umeå (48%, Po0.001). where À 0.083 ¼À0.118 þ 1.746*0.02 and 0.18 ¼ 0.134 þ 0.046 The exposure was higher for those who had occupational (Supplementary Table 7)). The exposure for persons from the exposure (16%), and for those with a gas-stove (42%), other cities and campaigns was estimated correspondingly. Supplementary Table 5. The exposure was lower for the 5% of the subjects with oil heating in their single-family houses (24% 1,3-Butadiene. The results for all subjects showed that the 1,3- lower). The exposure decreased with a higher proportion of time butadiene exposure was significantly lower in Gothenburg spent at home (P ¼ 0.047): if two people are compared who spend compared with Umeå, as seen in Supplementary Table 4. There 42% or 50% of their time at home, respectively, the person with was no significant change over time in the two cities with the longer time has 3.4% lower exposure. measurements in different years (Umeå and Malmo¨). The results among non-smokers were similar to those among all Smoking was a significant determinant and the exposure was subjects (Supplementary Table 5), but no significant difference was 175% higher for smokers (Supplementary Table 4). The exposure found between Umeå and Stockholm, and no effect of occupational increased with the proportion of time spent outdoors: if two exposure was found. Non-smokers exposed to ETS had 22% higher people are compared who spend either 2% or 4% of their time exposure (among all subjects, no significant effect of smoking was

2 2 Table 4. Between (s^B)- and within (s^W)-subject variances in empty and ﬁnal models.

ga Empty model Final model Reduction in variance (%)

Pg Pg Pg Pg 2 2 2 2 sbh swh sbh swh Between-subject Total h¼1 h¼1 h¼1 h¼1 Benzene All 8 3.25 1.09 1.76 1.06 46 35 Non-smokers 8 2.86 1.07 1.35 1.09 53 38

1,3-Butadiene All 7 5.19 1.87 3.23 1.75 38 29 Non-smokers 7 4.27 1.95 3.07 1.83 28 21

NO2 All 6 1.62 0.16b 0.39 0.17b 76 45c Non-smokers 6 1.46 0.18b 0.36 0.18b 75 43c

Formaldehyde All 8 1.85 0.39 1.21 0.39 35 29 Non-smokers 8 1.83 0.40 1.18 0.40 36 29 ag is the number of measuring campaigns included for the speciﬁc exposure. b For NO2 the within variance is the same for all campaigns. P6 c 2 2 The total variance for empty and ﬁnal model was estimated as sBh þ 6*s^W. h¼1

found). Having a gas-stove increased the exposure by 37%. DISCUSSION In Table 3, the estimated NO2 concentrations (using the model This study is unique from our point of view, as it includes for all subjects, Supplementary Table 5) for a person without measurements of personal exposure of current carcinogens in occupational exposure from various cities and campaigns are numerous randomly chosen individuals (209–293 persons, presented for comparison. This person is assumed to spend 70% of depending on measured substance) and in five geographically the time at home and not have a gas stove or oil heating at home. spread cities within one country. The measurements are compar- able as we used a similar sampling strategy in each city. In Formaldehyde. The results for all subjects showed that the addition, mixed-effects modeling offer opportunities to identify formaldehyde exposure was significantly lower in Stockholm, determinants of exposure, variability, and exposure estimations for compared with Umeå, and significantly higher in Lindesberg and the general population. Gothenburg compared with Umeå, Supplementary Table 6. No If these measuring campaigns can be continued in each city, significant change over time was found. every fifth year, this will allow time trends to be analyzed in the The exposure was higher for those living in single-family houses future. Until now we only have a maximum of two measurement (34%) compared with apartments, Supplementary Table 6. The occasions, which makes further conclusions unreliable. Meteor- exposure was lower for smokers ( À 19%), compared with those ological influences on concentrations will likely vary between non-smokers who were not exposed to ETS. cities. More measurement occasions will also allow adjustments The results among non-smokers showed similar differences for weather conditions. An example of seasonal variation in between the cities as for the total group, Supplementary Table 6. temperature and NO2 concentration at the urban background For those cities with measurements in two different years (Umeå, station in Malmo¨ is given in Supplementary Figure1. Malmo¨, Gothenburg), a significant decrease over time was found in Gothenburg and Malmo¨ (11% and 31% lower, respectively). As Determinants of Exposure for all subjects, the exposure was higher for those living in single- Benzene. In our study, refueling gasoline increased personal family houses (34%) compared with apartments. The exposure exposure levels of benzene (Table 2). Proportion of time spent in was lower for non-smokers exposed to ETS ( À 10%), compared traffic did not significantly affect exposure to benzene for all with the other non-smokers. participants, however, for the sub-group of non-smokers, the In Table 3, the estimated formaldehyde concentrations (using proportion of time spent in traffic significantly affected the the model for all subjects, Supplementary Table 6) for a non- exposure (Table 2). A possible explanation may be that smoking smoker from various cities and campaigns are presented for blurred the expected impact of exposure to vehicle exhausts comparison. This person is assumed to live in a single-family when smokers were included in the study group. Cigarette house and is not exposed to ETS. smoking was a significant determinant for personal exposure to benzene, as it is a well-known source of benzene. Explained variances by the significant determinants in the models. An Australian study20 found results similar to ours. A mixed- In Table 4, estimates of the between-subject and the within- model analysis was performed to investigate various determinants, subject variances are presented for the final models as well as for adjusting (including) for season, sex, age, and studying the effect of models without any of the significant determinants (empty time spent commuting as well as refueling of a motor vehicle. For models). The determinants in the final models mainly reduced time spent in a motor vehicle, a significant contribution was the between-subject variance, but resulted in no, or very small, determined, and the corresponding significance was noted for reductions of the within-subject variances. Between 21% and 45% refueling. Each hour spent commuting resulted in a mean increase of the total variances were explained by the determinants in the of 0.74 mg/m3. Refueling increased the personal exposure of models. benzene by 1.50 mg/m3.

& 2014 Nature America, Inc. Journal of Exposure Science and Environmental Epidemiology (2014), 437 – 443 Determinants of personal exposure to carcinogenic substances and NO2 Hagenbjo¨rk-Gustafsson et al 442 The largest and the smallest city, Stockholm and Lindesberg, sampling. Another theory could be that smokers are more likely to respectively, showed significantly higher exposure levels. In open the windows compared with others, which would reduce the Stockholm, traffic emissions are high, and in Lindesberg, emis- indoor concentration of formaldehyde and pollutants originating sions of wood smoke might be a contributing factor because of from mainly indoor sources. frequent use of wood stoves in that area.7 Model Data Compared with Other Reported Studies 1,3-Butadiene. Our data indicates a significantly lower exposure Personal air sampling data reflecting the exposures for the general in Gothenburg compared with Umeå. The reason is unknown, but population are sparse. Benzene exposure has been determined by similar findings have been found previously in outdoor samples personal monitoring in an Australian study.20 Similar to our from these two cities when analyzed at the same laboratory in the average levels, the average levels during summer and winter were same measuring campaign.21 Smoking was a significant 1.76 and 1.86 mg/m3, respectively. In a European study (EXPOLIS) determinant for 1,3-butadiene exposure, as well as outdoor performed between 1996 and1998, the personal exposure levels activity, proportion of time spent in traffic, and exposure to ETS of benzene ranged from 2 to 3 mg/m3 in Helsinki, Basel, and for non-smokers. For non-smokers, the effect of being outdoors Oxford, whereas higher levels were observed in Prague and significantly affected exposure to 1,3-butadiene, suggesting an Athens (8 and 12 mg/m3, respectively)29 as well as in the United impact from the general environment in addition to smoking States.18 In a review of traffic-related air pollution in developing habits. This is in agreement with a study from Australia, where the countries, the average level for a bus driver in India was highest levels of 1,3-butadiene and benzene were observed in 527.3 mg/m3.30 In Rouen, France, 5 days of personal exposure locations affected by traffic pollution, cigarette smoke (e.g., inside showed average benzene levels of 10.3 mg/m3.31 cars, basement car parks, and night clubs), and in two homes In the Spanish study among pregnant women,25 the median heated with wood.22 In a UK study among 100 non-smoking personal NO exposure was 40 mg/m3, as compared with our persons,23 11% of the variability of personal exposure was 2 average level of 14.1 mg/m3. In a study from Antwerp, Belgium,32 explained from exposure to ETS, use of solvents, and parking the average personal exposure to NO was 41.4 mg/m3. vehicles in garages attached to a house. The highest personal 2 Exposure to 1,3-butadiene has only been measured in a few exposure (6.3 mg/m3) was found in a subject exposed to ETS. published studies14,33,34 because of the lack of validated measurement methods. Kim et al reported a median personal NO . Personal exposure levels were higher in Stockholm, Malmo¨, 2 exposure level of 0.4 mg/m3 among 12 non-smoking persons in the and Umeå than in Lindesberg, probably due to higher traffic United Kingdom, and a similar median indoor level in these density, including higher diesel exhaust exposure. In a US study by subject’s homes.33 These results are in agreement with ours. The Williams et al, 2012, persons exposed to ETS showed a significantly mean 1,3-butadiene concentration was, however, almost six times higher NO level, at least in winter.24 In our study, smokers did not 2 higher in samples related with ETS compared with samples have significantly increased exposure levels, but non-smokers non-related with ETS.33 Much higher median personal exposure exposed to ETS had. However, this increase in exposure was not as (3 mg/m3) and indoor levels (3 mg/m3) were found in Mexico City.35 distinct as in the US study. In the present study, we found that The formaldehyde levels are in agreement with studies from using a gas stove was a significant determinant for NO exposure 2 several countries including Sweden, Great Britain, Mexico, France, (Supplementary Table 5). This is consistent with the study by and Finland.17,31,35–37 Williams, in which the use of air conditioning, gas stoves, or gas heating contributed to significantly increased exposure concentrations of NO2. Air conditioning as well as the use of gas stoves is Sampling and Analysis rare in Sweden, but using either of these are known to influence We have used published, commercially available sampling and 25 NO2 levels. In a study from Spain, personal 48-h NO2 exposure analytical methods. All analyses were performed by laboratories was measured among 53 pregnant women. Four variables participating in the campaigns. Diffusive sampling offers both (outdoor and indoor NO2 concentrations, time spent outdoors, practical and principal advantages compared with active sam- and time using a gas cooker) accounted for 59% of the variation in pling, and repeated measurements are relatively easy to perform, NO2 exposure levels in a predictive model used for the personal thereby reducing overall uncertainty. One drawback of the time exposure. No effect of time exposed to ETS on exposure was series is the ongoing modification of sampling and analytical found by Valero et al, which was explained by possible pollution- methods over time. However, for all the methods used, cross valida- avoiding behavior in the pregnant women. In our study the tion for the methods has been performed, resulting in small differ- exposure was lower for those with oil heating. One possible ences between the different passive samplers (data not shown). explanation could be that oil heating is more frequently used in older houses situated in less traffic-exposed quarters of the town Estimated Personal Mean Exposure Compared with Standards with no district heating. The estimated arithmetic mean concentrations for the general population (20–50 years of age) in Sweden in the mid 2000s were Formaldehyde. Living in a single-family house increased the 1.95 mg/m3 for benzene, 0.56 mg/m3 for 1,3-butadiene, 19.4 mg/m3 exposure significantly, as seen in Supplementary Table 6. This is for formaldehyde, and 14.1 mg/m3 for NO . in accordance with a previous study,17 and could be caused by 2 Our modeled and measured data for butadiene are within emission from building materials, the age of the source materials, the range of recommended low-risk levels of 0.2–1.0 mg/m3 for ventilation rates, indoor temperature, or relative humidity.2 Swedish 1,3-butadiene.38 The estimated mean exposure for benzene apartments have higher ventilation rates than single-family (1.95 mg/m3) was higher than the recommended low-risk level of houses,26 which could explain the lower levels found in apartments. 1.7 mg/m3 set by the WHO39 and the Swedish recommended Other major sources of indoor formaldehyde are combustion low-risk level of 1.3 mg/m3 based on the same scientific data. processes such as smoking, burning of candles, and cooking.27 In a review of Salthammer et al,28 cigarette smoke is mentioned as an important combustion source of formaldehyde indoors in Mixed Model developed countries. However, in our study, smokers had a Using a mixed model for analysis of determinants and variability significant lower exposure. One possible explanation for this offers possibilities of a description of fixed and random effects finding is that formaldehyde vapor adsorbs to particles formed when repeated measurements for individuals are available. The by cigarette smoke and is therefore not detected by diffusive improvements of our modeling are studied by comparing the

Journal of Exposure Science and Environmental Epidemiology (2014), 437 – 443 & 2014 Nature America, Inc. Determinants of personal exposure to carcinogenic substances and NO2 Hagenbjo¨rk-Gustafsson et al 443 empty model to the final model including determinants of 13 Levin JO, Sunesson AL, Sundgren M, Strandberg B eds).Determination of benzene exposure, which showed a substantial reduction of the between- and 1,3- butadiene in workplace and ambient air using diffusive sampling and subject and total variance in our models. automatic thermal desorption. Fifth International Symposium on Modern Principles Using the models, the expected exposure for individuals with of Air Monitoring. Loen, Norway, 2005 (Abstract O-54, oral presentation by Levin). different exposure patterns can be presented. The model also 14 Modig L, Sunesson AL, Levin JO, Sundgren M, Hagenbjork-Gustafsson A, Forsberg offers the possibility of presenting estimated exposure data with B. Can NO2 be used to indicate ambient and personal levels of benzene and 1,3- butadiene in air? J Environ Monit 2004; 6: 957–962. contributions from activities not directly linked to the ambient air 15 Sundgren M. Personal communication, 17 June, 2013. quality. 16 Levin JO, Lindahl R, Andersson K. High-performance liquid-chromatographic determination of formaldehyde in air in the Ppb to Ppm Range using diffusive sampling and hydrazone formation. Environ Technol Lett 1988; 9: 1423–1430. CONCLUSIONS 17 Gustafson P, Barregard L, Lindahl R, Sallsten G. Formaldehyde levels in Sweden: A comprehensive survey of personal exposure levels of benzene, personal exposure, indoor, and outdoor concentrations. J Expo Anal Environ Epi- demiol 2005; 15: 252–260. 1,3-butadiene, formaldehyde, and NO2 has been performed in five 18 Rappaport SM, Kupper LL. Variability of environmental exposures to volatile Swedish cities, and a model based on data from 2002 to 2008 is organic compounds. J Expo Anal Environ Epidemiol 2004; 14: 92–107. presented. Modeled, as well as measured, data were in line with 19 Rappaport SM, Kupper LL. Quantitative Exposure Assessment. El Cerrito: California, international recommendations, except for benzene. Including USA, 2008. determinants such as smoking, exposure to ETS, occupational 20 Horton A, Murray F, Bulsara M, Hinwood A, Farrar D. Personal monitoring of exposure, time spent in traffic, and refueling enabled us to study benzene in Perth, Western Australia: the contribution of sources to non-industrial the influence of different exposure-related factors. Time spent in personal exposure. Atmos Environ 2006; 40: 2596–2606. traffic and refueling were significant determinants of benzene 21 Strandberg B, Sunesson AL, Sundgren M, Levin JO, Sallsten G, Barregard L. Field exposure. The between-subject variance dominated for 1,3- evaluation of two diffusive samplers and two adsorbent media to determine 1,3-butadiene and benzene levels in air. Atmos Environ 2006; 40: 7686–7695. butadiene and formaldehyde, whereas the between-city variance 22 Hinwood AL, Berko HN, Farrar D, Galbally IE, Weeks IA. Volatile organic dominated for NO2. For benzene, the between-subject and compounds in selected micro-environments. Chemosphere 2006; 63: 421–429. between-city variances were similar. 23 Delgado-Saborit JM, Aquilina NJ, Meddings C, Baker S, Harrison RM. Relationship of personal exposure to volatile organic compounds to home, work and fixed site outdoor concentrations. Sci Total Environ 2011; 409: 478–488. ACKNOWLEDGEMENTS 24 Williams R, Jones P, Croghan C, Thornburg J, Rodes C. The influence of human and environmental exposure factors on personal NO2 exposures. J Expo Sci Environ This project was funded by the Swedish Environmental Protection Agency. We thank Epidemiol 2012; 22: 109–115. the study participants and Associate Professor Kåre Eriksson for comments on the 25 Valero N, Aguilera I, Llop S, Esplugues A, de Nazelle A, Ballester F et al. Concentrations manuscript. and determinants of outdoor, indoor and personal nitrogen dioxide in pregnant women from two Spanish birth cohorts. Environ Int 2009; 35: 1196–1201. 26 Norleń U, Andersson K. Indoor climate of Swedish dwellings, Report ELIB 7. REFERENCES Swedish Institute for building construction, Ga¨vle, Sweden, 1993 (In Swedish). 1 HEI Panel on the Health Effects of Traffic-Related Air Pollution. Traffic-Related Air 27 Heinrich J. Influence of indoor factors in dwellings on the development of Pollution: A Critical Review of the Literature on Emissions, Exposure, and Health childhood asthma. Int J Hygiene Environ Health 2011; 214: 3–27. Effects. HEI Special Report 17. Health Effects Institute: Boston, MA, 2010. 28 Salthammer T, Mentese S, Marutzky R. Formaldehyde in the Indoor Environment. 2 IARC. Chemical agents and related occupations, volume 100 F. A review of human Chem Rev 2010; 110: 2536–2572. carcinogens.IARC Monographs on the Evaluation of Carcinogenic Risks to 29 Saarela K, Tirkkonen T, Laine-Ylijoki J, Jurvelin J, Nieuwenhuijsen MJ, Jantunen M. Humans. International Agency for Research on Cancer: Lyon, France, 2012. Exposure of population and microenvironmental distributions of volatile organic 3 IARC. Agents classified by the IARC Monographs 1–107. http://monographs. compound concentrations in the EXPOLIS study. Atmos Environ 2003; 37: iarc.fr/ENG/Classification/ClassificationsAlphaOrder.pdfupdated on 10 April 2013. 5563–5575. 4 Oudin A, Forsberg B, Stromgren M, Beelen R, Modig L. Impact of residential 30 Mukherjee AK, Bhattacharya SK, Ahmed S, Roy SK, Roychowdhury A, Sen S. mobility on exposure assessment in longitudinal air pollution studies: a sensitivity Exposure of drivers and conductors to noise, heat, dust and volatile organic analysis within the ESCAPE project. Sci World J 2012; 2012: 125818. compounds in the state transport special buses of Kolkata city. Transp Res Part D 5 Bellander T, Berglind N, Gustavsson P, Jonson T, Nyberg F, Pershagen G et al. Transp Environ 2003; 8: 11–19. Using geographic information systems to assess individual historical exposure to 31 Gonzalez-Flesca N, Cicolella A, Bates M, Bastin E. Pilot study of personal, indoor air pollution from traffic and house heating in Stockholm. Environ Health Perspect and outdoor exposure to benzene, formaldehyde and acetaldehyde. Environ Sci 2001; 109: 633–639. Pollut Res Int 1999; 6: 95–102. 6 Esplugues A, Ballester F, Estarlich M, Llop S, Fuentes V, Mantilla E et al. Indoor and 32 Stranger M, Potgieter-Vermaak SS, Van Grieken R. Particulate matter and gaseous outdoor concentrations and determinants of NO2 in a cohort of 1-year-old chil- pollutants in residences in Antwerp, Belgium. Sci Total Environ 2009; 407: dren in Valencia, Spain. Indoor Air 2010; 20: 213–223. 1182–1192. 7 Gustafson P, Barregard L, Strandberg B, Sallsten G. The impact of domestic wood 33 Kim YM, Harrad S, Harrison RM. Levels and sources of personal inhalation expo- burning on personal, indoor and outdoor levels of 1,3-butadiene, benzene, for- sure to volatile organic compounds. Environ Sci Technol 2002; 36: 5405–5410. maldehyde and acetaldehyde. J Environ Monit 2007; 9: 23–32. 34 Zhou J, You Y, Bai Z, Hu Y, Zhang J, Zhang N. Health risk assessment of personal 8 Gustafson P, Ostman C, Sallsten G. Indoor levels of polycyclic aromatic hydro- inhalation exposure to volatile organic compounds in Tianjin, China. Sci Total carbons in homes with or without wood burning for heating. Environ Sci Technol Environ 2011; 409: 452–459. 2008; 42: 5074–5080. 35 Serrano-Trespalacios PI, Ryan L, Spengler JD. Ambient, indoor and personal 9 Bellander T, Wichmann J, Lind T. Individual exposure to NO2 in relation to spatial exposure relationships of volatile organic compounds in Mexico City Metropolitan and temporal exposure indices in Stockholm, Sweden: the INDEX study. PloS One Area. J Expo Anal Environ Epidemiol 2004; 14(Suppl 1): S118–S132. 2012; 7: e39536. 36 Jurvelin J, Vartianinen M, Jantunen M, Pasanen P. Personal exposure levels 10 Hagenbjo¨rk-Gustafsson A, Lindahl R, Levin JO, Karlsson D. Validation of a diffusive and microenvironmental concentrations of formaldehyde and acetaldehyde sampler for NO2. J Environ Monit 1999; 1: 349–352. in the Helsinki metropolitan area, Finland. J Air Waste Manage 2001; 51: q17–24. 11 Hagenbjo¨rk-Gustafsson A, Lindahl R, Levin JO, Karlsson D. Validation of the 37 Raw GJ, Coward SKG, Brown VM, Crump DR. Exposure to air pollutants in English Willems badge diffusive sampler for nitrogen dioxide determinations in occu- homes. J Expo Anal Environ Epidemiol 2004; 14: S85–S94. pational environments. Analyst 2002; 127: 163–168. 38 Finnberg N, Gustavsson P, Ho¨gberg J, Johanson G, Sallsten G, Warholm M, Victorin 12 Hagenbjo¨rk-Gustafsson A, Tornevi A, Forsberg B, Eriksson K. Field validation of the K. Risk Assessment of 1,3-Butadiene. Institute of Environmental Medicine (IMM). Ogawa diffusive sampler for NO2 and NOx in a cold climate. J Environ Monit 2010; Karolinska Institute: Stockholm, Sweden, 2004, IMM report 1/2004. 12: 1315–1324. 39 WHO. Air Quality Guidelines for Europe, 2nd edn. WHO: Copenhagen, 2000.

Supplementary Information accompanies the paper on the Journal of Exposure Science and Environmental Epidemiology website (http:// www.nature.com/jes)

& 2014 Nature America, Inc. Journal of Exposure Science and Environmental Epidemiology (2014), 437 – 443